Comparison of the proportional hazard model and the accelerated failure model in the mixed cure model

Abstract

Traditional survival analysis models such as the Cox model and the accelerated failure time model (AFT) assume that all individuals will eventually experience specified endpoint events, such as recurrence or death. However, in recent years, with the Advancement of science and technology and the improvement of medical standards, in many clinical trials, there are some individuals who will not experience terminal events after treatment, that is, they will not relapse or die. The researchers believe that these individuals have been cured and call them long-term survivors. In this case, using the traditional Cox model and the AFT model will cause large errors and affect the judgment. Therefore, we consider applying a mixed healing model to the data. In the previous period, we have compared the model of proportional risk function and proportional risk mixed healing model and accelerated failure function model with accelerated failure mixed healing model. In this paper, we want to compare the predicted effects of the PHMC model and the AFTMC model. Methods: We use Monte Carlo simulations to generate data that satisfy and do not satisfy proportional assumptions. Using the consistency probability, the average square error of regression coefficient and 95% confidence interval to cover the original parameter as the evaluation index, the discriminant precision and fitting effect of the same data are compared. Result: For the survival data based on the assumption of proportional risk, the fitting effect of PHMC model is more accurate than that of AFTMC model. For the survival data based on the assumption that the proportional risk is not satisfied, the fitting effect of AFTMC model is better than that of PHMC model. Conclusion: The PHMC model is recommended for survival data based on the assumption of proportional risk assumptions. The AFTMC model is recommended for survival data based on the assumption that the proportional risk is not met.